Active Vascular Shape Models For Blood Vessel Segmentation

Vessel segmentation is a key component in vessel imaging systems. It is anecessary precondition for three-dimensional visualization, morphologicalmeasurements and other subsequent processing. Vessel segmentation is now appliedin many clinical tasks, including surgical navigation, interventional therapy andpathological tracking. The precise segmentation results can help doctors diagnosevascular diseases, determine treatment schemes. Vessel segmentation plays a keyrole in clinic treatment.This dissertation focuses on the vessel segmentation algorithms based on activecontour models and lucubrates the vessel targets enhancement algorithms withtensor analysis, the prior vessel shape extraction algorithm using active contourmodels and the three-dimension vessel active contour segmentation model usingshape constrains. The main work is as follows.A new vessel enhancement algorithm using orthogonal tensor invariant isproposed in this dissertation to enhance the vessel targets due to the enhancementresults of classical approaches are sensitive to the parameters and noise. Theclassical vessel enhancement algorithms use Hessian matrixes to construct avesselness function owing to Hessian matrixes can describe the local imageinformation. However, this approach is difficult to obtain the optimal parameters.The new approach proposed in this dissertation analyze the fractional anisotropy ofHessian matrixes to construct a vesselness function according to the image Hessianmatrixes are equivalent to the mechanical stress tensors in theory. The vesselnessfunction is constructed to recognize the vessel targets in medical images and obtainthe shape of vessels. This new approach using mechanical stress tensor analysis has less parameters than the Frangi vessel enhancement algorithm and stronger adaptiveprocessing capacity.A new vessel segmentation algorithm with priori shape constraints, based onthe active contour model, is proposed in this dissertation due to the classicalsegmentation algorithms can not handle medical images with intensityinhomogeneity, which would cause a leakage problem in weak object boundaries.The new segmentation algorithm uses minimum variance energy item in energyfunctional to guarantee the global optimality and improve the anti-noise capability.The gradient vector field and the evolution contour alignment item are used toimprove the segmentation accuracy in fuzzy image. The vesselness function is usedto construct the vesselness prior shape knowledge, which can guide a correctevolution for active contour models. This approach could obtain a correct andprecisely segmentation result for images with intensity inhomogeneity andvesselness targets with weak edgeA three-dimension vessel segmentation algorithm using smaller principalcurvature and tubular constraints is proposed in this dissertation due to the classicalthree-dimension segmentation approach can not handle medical images with lowcontrast and narrow vessels. Tow orthogonal tensor invariant, fractional anisotropyand the mode of anisotropy of the image Hessian matrix, are used to construct athree-dimension vesselness function. The surface of the vessel prior shape, obtainedby the threshold of the vesselness function, is expressed implicitly by level setsigned distance. The surface smoothness is achieved using the smaller principalcurvature, which can smooth the surface of the vessel without changing its shape.The experiments show that our algorithm is more accurate and robust than theseclassical active contour models and is, therefore, more suited for automatic vesselsegmentation.